Add contr_one_hot function for one-hot encoding support

[EmilHvitfeldt]

Summary

The convert_data.R file references contr_one_hot in local_one_hot_contrasts() (line 142), but this function is not defined in tidyclust. This means the indicators = "one_hot" code path in .convert_form_to_x_fit() and .convert_form_to_x_new() is currently broken.

Current behavior

data <- data.frame(x = 1:10, cat = factor(rep(c("a", "b"), 5)))
.convert_form_to_x_fit(~., data, indicators = "one_hot")
#> Error: object 'contr_one_hot' of mode 'function' was not found

Options

1. Import from hardhat: The contr_one_hot function exists in the hardhat package. We could import it.

2. Define locally: Copy/adapt the function into tidyclust.

3. Remove the code path: If no engine uses indicators = "one_hot", consider removing this dead code.

Context

• Currently no engine in tidyclust uses predictor_indicators = "one_hot" encoding
• Tests for one-hot paths are currently skipped
• This was discovered during test coverage improvements

Related code

# R/convert_data.R lines 140-145
local_one_hot_contrasts <- function(frame = rlang::caller_env()) {
contrasts <- getOption("contrasts")
  contrasts["unordered"] <- "contr_one_hot"  # <- function doesn't exist
  rlang::local_options(contrasts = contrasts, .frame = frame)
}